csr_interface.c 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421
  1. /* StarPU --- Runtime system for heterogeneous multicore architectures.
  2. *
  3. * Copyright (C) 2008-2020 Université de Bordeaux, CNRS (LaBRI UMR 5800), Inria
  4. * Copyright (C) 2010 Mehdi Juhoor
  5. *
  6. * StarPU is free software; you can redistribute it and/or modify
  7. * it under the terms of the GNU Lesser General Public License as published by
  8. * the Free Software Foundation; either version 2.1 of the License, or (at
  9. * your option) any later version.
  10. *
  11. * StarPU is distributed in the hope that it will be useful, but
  12. * WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  14. *
  15. * See the GNU Lesser General Public License in COPYING.LGPL for more details.
  16. */
  17. #include <starpu.h>
  18. static int copy_any_to_any(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data);
  19. static const struct starpu_data_copy_methods csr_copy_data_methods_s =
  20. {
  21. .any_to_any = copy_any_to_any,
  22. };
  23. static void register_csr_handle(starpu_data_handle_t handle, unsigned home_node, void *data_interface);
  24. static int csr_pointer_is_inside(void *data_interface, unsigned node, void *ptr);
  25. static starpu_ssize_t allocate_csr_buffer_on_node(void *data_interface_, unsigned dst_node);
  26. static void free_csr_buffer_on_node(void *data_interface, unsigned node);
  27. static size_t csr_interface_get_size(starpu_data_handle_t handle);
  28. static int csr_compare(void *data_interface_a, void *data_interface_b);
  29. static uint32_t footprint_csr_interface_crc32(starpu_data_handle_t handle);
  30. static starpu_ssize_t describe(void *data_interface, char *buf, size_t size);
  31. static int pack_data(starpu_data_handle_t handle, unsigned node, void **ptr, starpu_ssize_t *count);
  32. static int unpack_data(starpu_data_handle_t handle, unsigned node, void *ptr, size_t count);
  33. struct starpu_data_interface_ops starpu_interface_csr_ops =
  34. {
  35. .register_data_handle = register_csr_handle,
  36. .allocate_data_on_node = allocate_csr_buffer_on_node,
  37. .free_data_on_node = free_csr_buffer_on_node,
  38. .copy_methods = &csr_copy_data_methods_s,
  39. .get_size = csr_interface_get_size,
  40. .interfaceid = STARPU_CSR_INTERFACE_ID,
  41. .interface_size = sizeof(struct starpu_csr_interface),
  42. .footprint = footprint_csr_interface_crc32,
  43. .compare = csr_compare,
  44. .describe = describe,
  45. .pointer_is_inside = csr_pointer_is_inside,
  46. .name = "STARPU_CSR_INTERFACE",
  47. .pack_data = pack_data,
  48. .unpack_data = unpack_data
  49. };
  50. static int csr_pointer_is_inside(void *data_interface, unsigned node, void *ptr)
  51. {
  52. (void) node;
  53. struct starpu_csr_interface *csr_interface = data_interface;
  54. return ((char*) ptr >= (char*) csr_interface->nzval &&
  55. (char*) ptr < (char*) csr_interface->nzval + csr_interface->nnz*csr_interface->elemsize)
  56. || ((char*) ptr >= (char*) csr_interface->colind &&
  57. (char*) ptr < (char*) csr_interface->colind + csr_interface->nnz*sizeof(uint32_t))
  58. || ((char*) ptr >= (char*) csr_interface->rowptr &&
  59. (char*) ptr < (char*) csr_interface->rowptr + (csr_interface->nrow+1)*sizeof(uint32_t));
  60. }
  61. static void register_csr_handle(starpu_data_handle_t handle, unsigned home_node, void *data_interface)
  62. {
  63. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *) data_interface;
  64. unsigned node;
  65. for (node = 0; node < STARPU_MAXNODES; node++)
  66. {
  67. struct starpu_csr_interface *local_interface = (struct starpu_csr_interface *)
  68. starpu_data_get_interface_on_node(handle, node);
  69. if (node == home_node)
  70. {
  71. local_interface->nzval = csr_interface->nzval;
  72. local_interface->colind = csr_interface->colind;
  73. }
  74. else
  75. {
  76. local_interface->nzval = 0;
  77. local_interface->colind = NULL;
  78. }
  79. local_interface->id = csr_interface->id;
  80. local_interface->rowptr = csr_interface->rowptr;
  81. local_interface->nnz = csr_interface->nnz;
  82. local_interface->nrow = csr_interface->nrow;
  83. local_interface->firstentry = csr_interface->firstentry;
  84. local_interface->elemsize = csr_interface->elemsize;
  85. }
  86. }
  87. /* declare a new data with the BLAS interface */
  88. void starpu_csr_data_register(starpu_data_handle_t *handleptr, int home_node,
  89. uint32_t nnz, uint32_t nrow, uintptr_t nzval, uint32_t *colind, uint32_t *rowptr, uint32_t firstentry, size_t elemsize)
  90. {
  91. struct starpu_csr_interface csr_interface =
  92. {
  93. .id = STARPU_CSR_INTERFACE_ID,
  94. .nnz = nnz,
  95. .nrow = nrow,
  96. .nzval = nzval,
  97. .colind = colind,
  98. .rowptr = rowptr,
  99. .firstentry = firstentry,
  100. .elemsize = elemsize
  101. };
  102. #ifndef STARPU_SIMGRID
  103. if (home_node >= 0 && starpu_node_get_kind(home_node) == STARPU_CPU_RAM)
  104. {
  105. if (nnz)
  106. {
  107. STARPU_ASSERT_ACCESSIBLE(nzval);
  108. STARPU_ASSERT_ACCESSIBLE(nzval + nnz*elemsize - 1);
  109. STARPU_ASSERT_ACCESSIBLE(colind);
  110. STARPU_ASSERT_ACCESSIBLE((uintptr_t) colind + nnz*sizeof(uint32_t) - 1);
  111. }
  112. STARPU_ASSERT_ACCESSIBLE(rowptr);
  113. STARPU_ASSERT_ACCESSIBLE((uintptr_t) rowptr + (nrow+1)*sizeof(uint32_t) - 1);
  114. }
  115. #endif
  116. starpu_data_register(handleptr, home_node, &csr_interface, &starpu_interface_csr_ops);
  117. }
  118. static uint32_t footprint_csr_interface_crc32(starpu_data_handle_t handle)
  119. {
  120. return starpu_hash_crc32c_be(starpu_csr_get_nnz(handle), 0);
  121. }
  122. static int csr_compare(void *data_interface_a, void *data_interface_b)
  123. {
  124. struct starpu_csr_interface *csr_a = (struct starpu_csr_interface *) data_interface_a;
  125. struct starpu_csr_interface *csr_b = (struct starpu_csr_interface *) data_interface_b;
  126. /* Two matricess are considered compatible if they have the same size */
  127. return (csr_a->nnz == csr_b->nnz)
  128. && (csr_a->nrow == csr_b->nrow)
  129. && (csr_a->elemsize == csr_b->elemsize);
  130. }
  131. /* offer an access to the data parameters */
  132. uint32_t starpu_csr_get_nnz(starpu_data_handle_t handle)
  133. {
  134. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  135. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  136. #ifdef STARPU_DEBUG
  137. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  138. #endif
  139. return csr_interface->nnz;
  140. }
  141. uint32_t starpu_csr_get_nrow(starpu_data_handle_t handle)
  142. {
  143. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  144. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  145. #ifdef STARPU_DEBUG
  146. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  147. #endif
  148. return csr_interface->nrow;
  149. }
  150. uint32_t starpu_csr_get_firstentry(starpu_data_handle_t handle)
  151. {
  152. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  153. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  154. #ifdef STARPU_DEBUG
  155. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  156. #endif
  157. return csr_interface->firstentry;
  158. }
  159. size_t starpu_csr_get_elemsize(starpu_data_handle_t handle)
  160. {
  161. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  162. starpu_data_get_interface_on_node(handle, STARPU_MAIN_RAM);
  163. #ifdef STARPU_DEBUG
  164. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  165. #endif
  166. return csr_interface->elemsize;
  167. }
  168. uintptr_t starpu_csr_get_local_nzval(starpu_data_handle_t handle)
  169. {
  170. unsigned node;
  171. node = starpu_worker_get_local_memory_node();
  172. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  173. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  174. starpu_data_get_interface_on_node(handle, node);
  175. #ifdef STARPU_DEBUG
  176. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  177. #endif
  178. return csr_interface->nzval;
  179. }
  180. uint32_t *starpu_csr_get_local_colind(starpu_data_handle_t handle)
  181. {
  182. unsigned node;
  183. node = starpu_worker_get_local_memory_node();
  184. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  185. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  186. starpu_data_get_interface_on_node(handle, node);
  187. #ifdef STARPU_DEBUG
  188. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  189. #endif
  190. return csr_interface->colind;
  191. }
  192. uint32_t *starpu_csr_get_local_rowptr(starpu_data_handle_t handle)
  193. {
  194. unsigned node;
  195. node = starpu_worker_get_local_memory_node();
  196. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  197. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *)
  198. starpu_data_get_interface_on_node(handle, node);
  199. #ifdef STARPU_DEBUG
  200. STARPU_ASSERT_MSG(csr_interface->id == STARPU_CSR_INTERFACE_ID, "Error. The given data is not a csr.");
  201. #endif
  202. return csr_interface->rowptr;
  203. }
  204. static size_t csr_interface_get_size(starpu_data_handle_t handle)
  205. {
  206. size_t size;
  207. uint32_t nnz = starpu_csr_get_nnz(handle);
  208. uint32_t nrow = starpu_csr_get_nrow(handle);
  209. size_t elemsize = starpu_csr_get_elemsize(handle);
  210. size = nnz*elemsize + nnz*sizeof(uint32_t) + (nrow+1)*sizeof(uint32_t);
  211. return size;
  212. }
  213. /* memory allocation/deallocation primitives for the BLAS interface */
  214. /* returns the size of the allocated area */
  215. static starpu_ssize_t allocate_csr_buffer_on_node(void *data_interface_, unsigned dst_node)
  216. {
  217. uintptr_t addr_nzval = 0;
  218. uint32_t *addr_colind = NULL, *addr_rowptr = NULL;
  219. starpu_ssize_t allocated_memory;
  220. /* we need the 3 arrays to be allocated */
  221. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *) data_interface_;
  222. uint32_t nnz = csr_interface->nnz;
  223. uint32_t nrow = csr_interface->nrow;
  224. size_t elemsize = csr_interface->elemsize;
  225. if (nnz)
  226. {
  227. addr_nzval = starpu_malloc_on_node(dst_node, nnz*elemsize);
  228. if (!addr_nzval)
  229. goto fail_nzval;
  230. addr_colind = (uint32_t*) starpu_malloc_on_node(dst_node, nnz*sizeof(uint32_t));
  231. if (!addr_colind)
  232. goto fail_colind;
  233. }
  234. else
  235. {
  236. addr_nzval = addr_colind = NULL;
  237. }
  238. addr_rowptr = (uint32_t*) starpu_malloc_on_node(dst_node, (nrow+1)*sizeof(uint32_t));
  239. if (!addr_rowptr)
  240. goto fail_rowptr;
  241. /* allocation succeeded */
  242. allocated_memory =
  243. nnz*elemsize + nnz*sizeof(uint32_t) + (nrow+1)*sizeof(uint32_t);
  244. /* update the data properly in consequence */
  245. csr_interface->nzval = addr_nzval;
  246. csr_interface->colind = addr_colind;
  247. csr_interface->rowptr = addr_rowptr;
  248. return allocated_memory;
  249. fail_rowptr:
  250. if (nnz)
  251. starpu_free_on_node(dst_node, (uintptr_t) addr_colind, nnz*sizeof(uint32_t));
  252. fail_colind:
  253. if (nnz)
  254. starpu_free_on_node(dst_node, addr_nzval, nnz*elemsize);
  255. fail_nzval:
  256. /* allocation failed */
  257. return -ENOMEM;
  258. }
  259. static void free_csr_buffer_on_node(void *data_interface, unsigned node)
  260. {
  261. struct starpu_csr_interface *csr_interface = (struct starpu_csr_interface *) data_interface;
  262. uint32_t nnz = csr_interface->nnz;
  263. uint32_t nrow = csr_interface->nrow;
  264. size_t elemsize = csr_interface->elemsize;
  265. if (nnz)
  266. {
  267. starpu_free_on_node(node, csr_interface->nzval, nnz*elemsize);
  268. starpu_free_on_node(node, (uintptr_t) csr_interface->colind, nnz*sizeof(uint32_t));
  269. }
  270. starpu_free_on_node(node, (uintptr_t) csr_interface->rowptr, (nrow+1)*sizeof(uint32_t));
  271. }
  272. /* as not all platform easily have a BLAS lib installed ... */
  273. static int copy_any_to_any(void *src_interface, unsigned src_node, void *dst_interface, unsigned dst_node, void *async_data)
  274. {
  275. struct starpu_csr_interface *src_csr = (struct starpu_csr_interface *) src_interface;
  276. struct starpu_csr_interface *dst_csr = (struct starpu_csr_interface *) dst_interface;
  277. uint32_t nnz = src_csr->nnz;
  278. uint32_t nrow = src_csr->nrow;
  279. size_t elemsize = src_csr->elemsize;
  280. int ret = 0;
  281. if (nnz)
  282. {
  283. if (starpu_interface_copy(src_csr->nzval, 0, src_node, dst_csr->nzval, 0, dst_node, nnz*elemsize, async_data))
  284. ret = -EAGAIN;
  285. if (starpu_interface_copy((uintptr_t)src_csr->colind, 0, src_node, (uintptr_t)dst_csr->colind, 0, dst_node, nnz*sizeof(uint32_t), async_data))
  286. ret = -EAGAIN;
  287. }
  288. if (starpu_interface_copy((uintptr_t)src_csr->rowptr, 0, src_node, (uintptr_t)dst_csr->rowptr, 0, dst_node, (nrow+1)*sizeof(uint32_t), async_data))
  289. ret = -EAGAIN;
  290. starpu_interface_data_copy(src_node, dst_node, nnz*elemsize + (nnz+nrow+1)*sizeof(uint32_t));
  291. return ret;
  292. }
  293. static starpu_ssize_t describe(void *data_interface, char *buf, size_t size)
  294. {
  295. struct starpu_csr_interface *csr = (struct starpu_csr_interface *) data_interface;
  296. return snprintf(buf, size, "C%ux%ux%u",
  297. (unsigned) csr->nnz,
  298. (unsigned) csr->nrow,
  299. (unsigned) csr->elemsize);
  300. }
  301. static int pack_data(starpu_data_handle_t handle, unsigned node, void **ptr, starpu_ssize_t *count)
  302. {
  303. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  304. struct starpu_csr_interface *csr = (struct starpu_csr_interface *) starpu_data_get_interface_on_node(handle, node);
  305. // We first pack colind
  306. *count = csr->nnz * sizeof(csr->colind[0]);
  307. // Then rowptr
  308. *count += (csr->nrow + 1) * sizeof(csr->rowptr[0]);
  309. // Then nnzval
  310. *count += csr->nnz * csr->elemsize;
  311. if (ptr != NULL)
  312. {
  313. *ptr = (void *)starpu_malloc_on_node_flags(node, *count, 0);
  314. char *tmp = *ptr;
  315. if (nnz)
  316. {
  317. memcpy(tmp, (void*)csr->colind, csr->nnz * sizeof(csr->colind[0]));
  318. tmp += csr->nnz * sizeof(csr->colind[0]);
  319. memcpy(tmp, (void*)csr->rowptr, (csr->nrow + 1) * sizeof(csr->rowptr[0]));
  320. tmp += (csr->nrow + 1) * sizeof(csr->rowptr[0]);
  321. }
  322. memcpy(tmp, (void*)csr->nzval, csr->nnz * csr->elemsize);
  323. }
  324. return 0;
  325. }
  326. static int unpack_data(starpu_data_handle_t handle, unsigned node, void *ptr, size_t count)
  327. {
  328. STARPU_ASSERT(starpu_data_test_if_allocated_on_node(handle, node));
  329. struct starpu_csr_interface *csr = (struct starpu_csr_interface *) starpu_data_get_interface_on_node(handle, node);
  330. STARPU_ASSERT(count == (csr->nnz * sizeof(csr->colind[0]))+((csr->nrow + 1) * sizeof(csr->rowptr[0]))+(csr->nnz * csr->elemsize));
  331. char *tmp = ptr;
  332. if (nnz)
  333. {
  334. memcpy((void*)csr->colind, tmp, csr->nnz * sizeof(csr->colind[0]));
  335. tmp += csr->nnz * sizeof(csr->colind[0]);
  336. memcpy((void*)csr->rowptr, tmp, (csr->nrow + 1) * sizeof(csr->rowptr[0]));
  337. tmp += (csr->nrow + 1) * sizeof(csr->rowptr[0]);
  338. }
  339. memcpy((void*)csr->nzval, tmp, csr->nnz * csr->elemsize);
  340. starpu_free_on_node_flags(node, (uintptr_t)ptr, count, 0);
  341. return 0;
  342. }